[CI]【Hackathon 10th Spring No.32】improve load_weight_utils coverage to 83%

- Add test_load_ep_checkpoint_basic: exercises EP checkpoint loading with minimal fixture
- Add test_composite_ep_branch: covers EP path in load_composite_checkpoint
- Add test_get_weight_iterator_unordered: covers unordered sharded safetensors path
- Coverage: 66% → 83% (257/310 stmts)

Author: cloudforge1

tests/model_executor/test_load_weight_utils.py

@@ -21,6 +21,7 @@
 
 import numpy as np
 import paddle
+import pytest
 from safetensors.numpy import save_file
 
 from fastdeploy.model_executor.load_weight_utils import (
@@ -31,6 +32,7 @@
     is_weight_cache_enabled,
     kv_cache_scale_iterator,
     load_composite_checkpoint,
+    load_ep_checkpoint,
     load_kv_cache_scale,
     load_pre_sharded_checkpoint,
     load_weights_from_cache,
@@ -458,3 +460,60 @@ def test_composite_pre_sharded(self, monkeypatch):
             mock_cls = SimpleNamespace(_get_tensor_parallel_mappings=lambda _: {})
             result = load_composite_checkpoint(d, mock_cls, cfg)
             assert "w" in result
+
+    # ── load_ep_checkpoint ─────────────────────────────────────────────
+
+    def test_load_ep_checkpoint_basic(self):
+        with tempfile.TemporaryDirectory() as d:
+            save_file(
+                {"w": np.array([1.0, 2.0], dtype=np.float32)},
+                os.path.join(d, "s1.safetensors"),
+            )
+            index = {"weight_map": {"w": "s1.safetensors"}}
+            with open(os.path.join(d, "model.safetensors.index.json"), "w") as f:
+                json.dump(index, f)
+            cfg = _make_fd_config()
+            cfg.parallel_config.num_experts_start_offset = 0
+            cfg.parallel_config.num_experts_per_rank = 1
+            cfg.model_config.moe_num_experts = 2
+            cfg.model_config.moe_layer_start_index = 0
+            cfg.model_config.num_hidden_layers = 1
+            cfg.speculative_config = SimpleNamespace(model_type="main")
+            cfg.parallel_config.use_sequence_parallel_moe = False
+            mock_cls = SimpleNamespace(_get_tensor_parallel_mappings=lambda _: {})
+            result = load_ep_checkpoint(mock_cls, d, cfg, return_numpy=True)
+            assert "w" in result
+            np.testing.assert_allclose(result["w"], [1.0, 2.0], rtol=1e-6)
+
+    def test_composite_ep_branch(self, monkeypatch):
+        cfg = _make_fd_config()
+        cfg.parallel_config.use_ep = True
+        cfg.quant_config.kv_cache_quant_type = "none"
+        monkeypatch.setattr(
+            "fastdeploy.model_executor.load_weight_utils.load_ep_checkpoint",
+            lambda cls, path, fd_config, return_numpy=True: {"w": np.zeros((2,))},
+        )
+        mock_cls = SimpleNamespace(_get_tensor_parallel_mappings=lambda _: {})
+        with tempfile.TemporaryDirectory() as d:
+            result = load_composite_checkpoint(d, mock_cls, cfg)
+        assert "w" in result
+
+    # ── get_weight_iterator (unordered sharded) ────────────────────────
+
+    def test_get_weight_iterator_unordered(self):
+        with tempfile.TemporaryDirectory() as d:
+            path = os.path.join(d, "model-001.safetensors")
+            save_file(
+                {"z_last": np.array([1.0], dtype=np.float32), "a_first": np.array([2.0], dtype=np.float32)},
+                path,
+            )
+            # Keys deliberately not naturally sorted → triggers ordered iterator path
+            index = {"weight_map": {"z_last": "model-001.safetensors", "a_first": "model-001.safetensors"}}
+            with open(os.path.join(d, "model.safetensors.index.json"), "w") as f:
+                json.dump(index, f)
+            results = dict(get_weight_iterator(d))
+            assert "z_last" in results and "a_first" in results
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])